A reproducible framework for constructing a longitudinal low-dose CT screening image database: implementation using...

Area of Science:

• Medical Imaging
• Radiology
• Data Science

Background:

• Longitudinal imaging data is crucial for screening studies.
• Standardized frameworks are needed for managing large-scale screening databases.
• Previous efforts lacked a unified, reproducible methodology for longitudinal LDCT data.

Purpose of the Study:

• To present a reproducible methodological framework for constructing a longitudinal low-dose computed tomography (LDCT) screening image database.
• To demonstrate the framework's implementation using real-world data.
• To enable consistent organization and analysis of longitudinal imaging data.

Main Methods:

• Developed a framework integrating hierarchical identifier design, deterministic metadata linkage, and structured anonymization.
• Implemented the framework using 11 years of LDCT data from Japan.
• Linked imaging data with screening assessment categories and smoking exposure information.

Main Results:

• Created a database of 45,337 LDCT examinations from 23,065 examinees, with 47.0% undergoing repeated screening.
• The implemented structure supports reproducible subject-level aggregation and temporal tracking.
• Demonstrated the framework's capability for quantitative image analysis.

Conclusions:

• The proposed framework provides a reproducible and transferable model for building longitudinal screening imaging repositories.
• This methodology facilitates consistent organization and analysis of large-scale LDCT screening data.
• The database supports advanced research in lung cancer screening and related fields.